A novel approach for classifying continuous speech into visible mouth-shape related classes.